The ANTsPyMM system takes advantage of template-based
priors.
We quantify cortex with the Desikan-Killiany-Tourville parcellation (DKT).
Subcortical segmentation with deep learning and also SyN registration leverages the CIT168 atlas 10.1101/211201.
A medial temporal lobe parcellation is based on manual labels derived from M. Yassa’s research group (preprint).
A cerebellum parcellation based on a template from the CoBrALab with
additional manual editing by the Tustison family (publication in
progress). The parcellation scheme is defined using the Schmahmann
nomenclature (reviewed
here). The core tool is named cerebellum_morphology and
based on a concatentation of two U-nets.
Brain stem is subdivided into the mid-brain, pons and medulla; This segmentation derives from registration between the individual T1w and the manually labeled CIT168 template.
We segment anatomy related to the basal forebrain and associated
cholinergic neurons with a deep learning method (deepNBM)
derived from manual labeling of ADNI data.
This approach is based on anatomical landmarks proposed in Liu, A. et
al. (2015). The deepNBM method uses super-resolution
segmentation to take advantage of manual labels performed on
super-resolution training data.
The classic Zaborszky, L. et al. (2008) “Stereotaxic probabilistic maps of the magnocellular cell groups in human basal forebrain.” Neuroimage 42: 1127–1141 also informed the manual labeling.
The Mori JHU white matter atlas provides a parcellation of white matter regions (paper here) which are applied to diffusion weighted images.
Jonathan Power’s coordinates from “Functional Network Organization of the Human Brain” 10.1016/j.neuron.2011.09.006 guide our analysis of resting state function MRI (rsfMRI).
The ANTsPyMM data dictionary labels associates each
column name with each of the above anatomical references. Table 1 below
summarizes the number of variables associated with each modality and
anatomical prior space. DTI-derived connectivity yields the largest
number of variables because we pairwise map each DKT cortical region as
well as the primary CIT168 regions.
| DTI | Flair | Neuromelanin | Other | restingStatefMRI | T1 DiReCT thickness processing | T1 hierarchical processing | unique id | p | |
|---|---|---|---|---|---|---|---|---|---|
| (N=9814) | (N=9) | (N=34) | (N=95) | (N=631) | (N=64) | (N=854) | (N=6) | ||
| Atlas | < 0.001 | ||||||||
| ANTs | 15 ( 0.2%) | 8 (100.0%) | 17 (51.5%) | 95 (100.0%) | 0 ( 0.0%) | 1 ( 1.6%) | 21 ( 2.5%) | 0 ( 0.0%) | |
| BF | 32 ( 0.3%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 24 ( 2.8%) | 0 ( 0.0%) | |
| CIT168 | 3464 (35.3%) | 0 ( 0.0%) | 16 (48.5%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 189 (22.2%) | 0 ( 0.0%) | |
| desikan-killiany-tourville | 6204 (63.2%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 28 ( 4.4%) | 62 (98.4%) | 486 (57.0%) | 0 ( 0.0%) | |
| johns hopkins white matter | 98 ( 1.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | |
| MTL | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 36 ( 4.2%) | 0 ( 0.0%) | |
| power peterson fMRI meta-analyses | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 602 (95.6%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | |
| quality control metrics | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 25 ( 2.9%) | 0 ( 0.0%) | |
| TustisonCobra | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 72 ( 8.4%) | 0 ( 0.0%) | |
| unique id | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 0 ( 0.0%) | 6 (100.0%) | |
| Measurement | unique values:34 | unique values:34 | unique values:34 | unique values:34 | unique values:34 | unique values:34 | unique values:34 | unique values:34 | |
Automated (blind) QC is part of standard ANTsPyMM processing. QC metrics are therefore present in the standard output tables. Notes on the meaning of these metrics, along with examples of the metrics on images, are here.
The outlierness function produces additional blind QC
measurements via automated outlierness calculations. These measurements
are denoted by Local Outlier Probability (LOOP *loop* ) and
local outlier factor ( LOF *lof* ) column names.
Time series images may have additional QC metrics such as:
TSNR: temporal signal to noise ratio
SSNR: spatial signal to noise ratio on the mean image (different from TSNR)
DVARS: the spatial root mean square of the data after temporal differencing
Neuromelanin (NM) has its own set of variables including:
count: the number of acquisitions at that date
min/max/mean/sd: a statistic on the raw image signal after averaging the acquisitions
avg/std_refregion : average and standard deviation
of signal in the reference region
avg/std_substantianigra : average and standard
deviation of signal in the substantia nigra or a related CIT168 region
(determined by deepCIT168)
NM2DMT_NM_substantianigra_z_coordinate: estimate of
the normalized (zero to one) z-coordinate of the substantia nigra in a
neuromelanin scan; higher values mean higher in the slab.
the framewise displacement (FD) values are 10x in DTI data compared to rsfMRI data
A sortable version of the data dictionary is below (if viewing on local machine).
We provide an example of how to join derived ANTsPyMM
summary data with clinical variables from PPMI.
FIXME
## [1] "T1Hier_thk_LRAVG_ca1mtl"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -2.607 9.67e-03
## DXSubAsynPDLRRK2Positive -4.054 6.54e-05
## yearsbl:DXSubAsynPDLRRK2Negative 0.889 3.74e-01
## yearsbl:DXSubAsynPDLRRK2Positive 1.056 2.91e-01
## [1] "DTI_mean_md_Asym_rostral_middle_frontal"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -1.3139 1.91e-01
## DXSubAsynPDLRRK2Positive -0.0959 9.24e-01
## yearsbl:DXSubAsynPDLRRK2Negative 4.6963 3.76e-06
## yearsbl:DXSubAsynPDLRRK2Positive -0.1794 8.58e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## [1] "T1Hier_thk_mtg_sn_snc_LRAVGdeep_cit168"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -3.373 8.55e-04
## DXSubAsynPDLRRK2Positive -4.670 4.62e-06
## yearsbl:DXSubAsynPDLRRK2Negative -1.181 2.38e-01
## yearsbl:DXSubAsynPDLRRK2Positive -0.365 7.15e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -3.0e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -6.2e+02
## [1] "T1Hier_thk_Asym_subiculummtl"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -3.900 1.22e-04
## DXSubAsynPDLRRK2Positive -4.878 1.79e-06
## yearsbl:DXSubAsynPDLRRK2Negative -0.267 7.90e-01
## yearsbl:DXSubAsynPDLRRK2Positive 1.204 2.29e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -1.3e+02
## [1] "T1Hier_thk_mtg_vtr_pbp_Asymcit168"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 1.978 4.90e-02
## DXSubAsynPDLRRK2Positive 4.153 4.35e-05
## yearsbl:DXSubAsynPDLRRK2Negative 2.638 8.50e-03
## yearsbl:DXSubAsynPDLRRK2Positive 0.444 6.57e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -5.3e+02
## Warning: Model failed to converge with 1 negative eigenvalue: -2.5e+02
## [1] "DTI_mean_md_LRAVG_inferior_temporal"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -2.46 1.55e-02
## DXSubAsynPDLRRK2Positive -2.68 8.52e-03
## yearsbl:DXSubAsynPDLRRK2Negative 4.95 1.13e-06
## yearsbl:DXSubAsynPDLRRK2Positive -1.15 2.50e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## boundary (singular) fit: see help('isSingular')
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## [1] "T1Hier_thk_LRAVG_pars_triangularisdktcortex"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 4.51 9.82e-06
## DXSubAsynPDLRRK2Positive 5.46 1.02e-07
## yearsbl:DXSubAsynPDLRRK2Negative -1.13 2.60e-01
## yearsbl:DXSubAsynPDLRRK2Positive 1.06 2.91e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -7.7e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -4.3e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -5.5e+06
## Warning: Model failed to converge with 1 negative eigenvalue: -3.9e+02
## [1] "T1Hier_thk_LRAVG_superior_frontaldktcortex"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 2.621 9.31e-03
## DXSubAsynPDLRRK2Positive 4.212 3.47e-05
## yearsbl:DXSubAsynPDLRRK2Negative -0.953 3.41e-01
## yearsbl:DXSubAsynPDLRRK2Positive 2.324 2.04e-02
## Warning: Model failed to converge with 1 negative eigenvalue: -1.7e+02
## Warning: Model failed to converge with 1 negative eigenvalue: -1.2e+03
## Warning: Model failed to converge with 1 negative eigenvalue: -3.2e+02
## Warning: Model failed to converge with 1 negative eigenvalue: -5.3e+02
## [1] "T1Hier_thk_LRAVG_precentraldktregions"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 2.339 2.01e-02
## DXSubAsynPDLRRK2Positive 4.414 1.46e-05
## yearsbl:DXSubAsynPDLRRK2Negative -0.572 5.68e-01
## yearsbl:DXSubAsynPDLRRK2Positive 1.367 1.72e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -2.4e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -3.5e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -3.3e+02
## [1] "DTI_mean_md_LRAVG_parahippocampal"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -1.9976 4.84e-02
## DXSubAsynPDLRRK2Positive -0.6550 5.14e-01
## yearsbl:DXSubAsynPDLRRK2Negative 4.6186 5.50e-06
## yearsbl:DXSubAsynPDLRRK2Positive -0.0249 9.80e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## [1] "T1Hier_thk_mtg_sn_snc_LRAVGsnseg"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -3.604 3.75e-04
## DXSubAsynPDLRRK2Positive -5.034 8.53e-07
## yearsbl:DXSubAsynPDLRRK2Negative 0.813 4.16e-01
## yearsbl:DXSubAsynPDLRRK2Positive 0.382 7.03e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -1.1e+01
## [1] "DTI_mean_md_Asym_transverse_temporal"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 0.494 6.22e-01
## DXSubAsynPDLRRK2Positive -0.971 3.33e-01
## yearsbl:DXSubAsynPDLRRK2Negative 4.323 2.02e-05
## yearsbl:DXSubAsynPDLRRK2Positive 1.860 6.37e-02
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Model failed to converge with 1 negative eigenvalue: -3.4e+03
## [1] "T1Hier_vol_mtg_sn_snc_LRAVGdeep_cit168"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -4.143 4.62e-05
## DXSubAsynPDLRRK2Positive -4.875 1.80e-06
## yearsbl:DXSubAsynPDLRRK2Negative 0.508 6.11e-01
## yearsbl:DXSubAsynPDLRRK2Positive -0.462 6.44e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Model failed to converge with 1 negative eigenvalue: -8.0e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -2.1e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -1.5e+02
## [1] "DTI_mean_md_LRAVG_fusiform"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -1.5423 1.26e-01
## DXSubAsynPDLRRK2Positive -2.7412 7.13e-03
## yearsbl:DXSubAsynPDLRRK2Negative 4.5835 6.39e-06
## yearsbl:DXSubAsynPDLRRK2Positive 0.0768 9.39e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## [1] "DTI_mean_md_LRAVG_pars_triangularis"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative -2.293 2.37e-02
## DXSubAsynPDLRRK2Positive -2.020 4.57e-02
## yearsbl:DXSubAsynPDLRRK2Negative 4.189 3.56e-05
## yearsbl:DXSubAsynPDLRRK2Positive -0.838 4.02e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Model failed to converge with 1 negative eigenvalue: -1.2e+02
## Warning: Model failed to converge with 1 negative eigenvalue: -2.7e+05
## Warning: Model failed to converge with 1 negative eigenvalue: -4.3e+02
## Warning: Model failed to converge with 1 negative eigenvalue: -4.5e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -2.2e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -1.3e+02
## Warning: Model failed to converge with 1 negative eigenvalue: -3.2e+02
## [1] "DTI_mean_md_Asym_parahippocampal"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 0.310 7.57e-01
## DXSubAsynPDLRRK2Positive 1.375 1.72e-01
## yearsbl:DXSubAsynPDLRRK2Negative 4.371 1.63e-05
## yearsbl:DXSubAsynPDLRRK2Positive 0.281 7.79e-01
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## [1] "hvlt_discrimination"
## t value Pr(>|t|)
## age -2.19 0.038584
## SEX 1.32 0.200543
## yearsbl -1.48 0.142140
## DXSubAsynPDLRRK2Positive -3.15 0.002354
## testvol -3.23 0.001710
## yearsbl:DXSubAsynPDLRRK2Positive 1.94 0.054974
## yearsbl:testvol 2.60 0.010713
## DXSubAsynPDLRRK2Positive:testvol 3.42 0.000935
## yearsbl:DXSubAsynPDLRRK2Positive:testvol -2.71 0.007832
## Warning: Removed 24 rows containing non-finite values (`stat_smooth()`).
## Warning: Removed 24 rows containing non-finite values (`stat_cor()`).
## Warning: Removed 579 rows containing non-finite values (`stat_smooth()`).
## Warning: Removed 579 rows containing non-finite values (`stat_cor()`).
## Warning: Removed 583 rows containing non-finite values (`stat_smooth()`).
## Warning: Removed 583 rows containing non-finite values (`stat_cor()`).
## TableGrob (1 x 3) "arrange": 3 grobs
## z cells name grob
## 1 1 (1-1,1-1) arrange gtable[layout]
## 2 2 (1-1,2-2) arrange gtable[layout]
## 3 3 (1-1,3-3) arrange gtable[layout]
## Warning: Model failed to converge with 1 negative eigenvalue: -6.0e+01
## [1] "T1Hier_thk_LRAVG_superior_parietaldktcortex"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 3.285 1.17e-03
## DXSubAsynPDLRRK2Positive 4.812 2.48e-06
## yearsbl:DXSubAsynPDLRRK2Negative 0.478 6.33e-01
## yearsbl:DXSubAsynPDLRRK2Positive 2.315 2.09e-02
## Warning: Model failed to converge with 1 negative eigenvalue: -1.4e+01
## Warning: Model failed to converge with 1 negative eigenvalue: -2.7e+03
## Warning: Model failed to converge with 1 negative eigenvalue: -4.1e+02
## [1] "DTI_mean_md_Asym_fusiform"
## t value Pr(>|t|)
## DXSubAsynPDLRRK2Negative 0.663 5.08e-01
## DXSubAsynPDLRRK2Positive 1.383 1.67e-01
## yearsbl:DXSubAsynPDLRRK2Negative 4.047 5.22e-05
## yearsbl:DXSubAsynPDLRRK2Positive 1.447 1.48e-01
## Warning: Model failed to converge with 1 negative eigenvalue: -4.1e+02
## Warning: Some predictor variables are on very different scales: consider
## rescaling
## Warning: Some predictor variables are on very different scales: consider
## rescaling